import { ObjectQuerySourceSchema } from '@objectstack/spec/ai';import type { ObjectQuerySource } from '@objectstack/spec/ai';// Validate dataconst result = ObjectQuerySourceSchema.parse(data);
The machine name of the object the figures were read from
label
string
optional
The object's plural label, or its label when it declares no plural. Translated only when the request is in Chinese and a zh* translation pack names the object; otherwise the authored label, not the caller's UI locale. Absent when the producer sets none: a reader falls back to object
filter
any
optional
The caller's where, in the $ query dialect, exactly as the engine accepted it: machine names only, no field labels, date placeholders such as {today} as written. Absent when no condition applied
scope
'visible_to_you'
✅
The row scope the figures cover: only the records the caller may read, after sharing and row-level narrowing. Always present; this is its only value
groupBy
string[]
optional
The field names the figures were grouped by, in order
The row or group cut applied (a top-N read or a group cut). Present only together with orderBy
related
string[]
optional
The lookup paths read in the same call, dotted from the object (for example client.relationship_manager)
latestPer
{ groupBy: string; orderField: string }
optional
Present when the answer keeps only the latest row per group
records
integer
optional
How many records the figures cover — a count, never a list: the match total for a page read, the summed group counts for an aggregate or a chart, 1 for get_record. Absent when unknown, never guessed
line
string
✅
The finished one-line citation, in the request language. The producer keeps it to at most 280 characters; a reader does not refuse a longer one
ObjectStack